The maximum distance problem and band sequences (Q1090739)

Let m,n be integers such that \(1\leq m<n\). For \(1\leq j,k\leq n\) and \(| j-k| \leq m\), let \(a_{ij}\) be given complex numbers with \(a_{kj}=\bar a_{jk}\). The authors wish to determine the basis \(x_ 1,...,x_ n\) of an n-dimensional complex inner product space such that \(<x_ j,x_ k>=a_{jk}\) for \(| j-k| \leq m\) and such that the distance from \(x_ k\) to the linear span of \(x_ 1,...,x_{k-1}\) is maximal for \(2\leq k\leq n.\) Assume the obvious necessary condition that each Hermitian matrix \((a_{\lambda \mu}),\quad \lambda,\mu =j+1,...,j+m\) (0\(\leq j\leq n-m)\) be positive definite. Then the solution, expressed geometrically, amounts to the following. Suppose \(x_ 1,...,x_{k-1}\) have been determined and we wish to construct \(x_ k\). Apply the Gram-Schmidt process to \(x_{k- 1},...,x_ 1\) to get the orthonormal set \(y_{k-1},...,y_ 1\). Express \(x_ k\) as \(\sum^{k-1}_{1}\lambda_ iy_ i+u\), where u is orthogonal to the \(y_ i\). Now, prescribing the inner products \(<x_ k,x_ i>=a_{ki}\) for \(r\leq i\leq k\), where \(r=\max (1,k-m+1),\) is equivalent to prescribing \(\lambda_ r,...,\lambda_{k-1}\) and \(<x_ k,x_ k>\). Since \(<x_ k,x_ k>=\sum^{k-1}_{1}| \lambda_ i|^ 2+<u,u>,\) the solution to the problem is obtained by taking \(\lambda_ i=0\) for \(1\leq i<r\) and u as any vector orthogonal to the \(y_ i\) and of squared length \(<x_ k,x_ k>-\sum^{k-1}_{r}| \lambda_ i|^ 2.\). The authors express their results in matrix terms and go into the details of finding the solutions explicitly. They show that their maximum distance problem is equivalent to a related maximum volume problem.